evaluation

Harness vs model: what actually moves a coding agent's benchmark score

LLM-as-a-judge: trusting model-graded evals

Context rot: why long-context LLMs underuse their window